The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install and execute a non-standard Python package.\n
Evidence: The skill recommends running uv pip install "datacommons-client[Pandas]" in SKILL.md and pip install "datacommons-client[Pandas]" in references/getting_started.md.\n
Risk: The package name datacommons-client does not match the official datacommons library name found in the linked documentation from the Data Commons project (Google). This discrepancy is a potential indicator of a dependency confusion or typosquatting attack, where a malicious package with a similar name is used to execute code in the user's environment.\n- [EXTERNAL_DOWNLOADS]: The skill is configured to interact with external data sources for querying public statistics.\n
Evidence: The skill makes requests to datacommons.org and its subdomains for data retrieval and entity resolution.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles data from external APIs, introducing a vulnerability surface for indirect prompt injection.\n
Ingestion points: Statistical data and metadata are retrieved from external sources via methods like client.observation.fetch and client.node.fetch.\n
Boundary markers: The provided code examples do not include delimiters or specific instructions to help the agent distinguish between data content and behavioral instructions.\n
Capability inventory: The skill is capable of performing file system operations, such as saving results to a CSV file (comparison.to_csv('ca_cities_comparison.csv')).\n
Sanitization: No explicit validation or sanitization of the statistical data received from the external API is demonstrated in the skill's references.

datacommons-client